Apparatus, method, and computer program for identifying state of signal light, and controller

1. An apparatus for identifying the state of a signal light, comprising:

a processor configured to: input time series images into a first classifier to detect object regions each including a vehicle equipped with a signal light in the respective images, the first classifier having been trained to detect the vehicle;

chronologically input characteristics obtained from pixel values of the object regions detected in the respective images into a second classifier to calculate confidence scores of possible candidate states of the signal light of the vehicle, the second classifier having a recursive structure or performing a convolution operation in a temporal direction; and

identify the state of the signal light, based on the preceding state of the signal light, information indicating whether transitions between the candidate states are allowed, and the confidence scores of the respective candidate states;

wherein the information indicates candidate states to which the respective candidate states are allowed to change, and of the candidate states to which the preceding state of the signal light is allowed to change, the processor identifies the state of the signal light as a candidate state having a maximum confidence score; and

wherein the candidate states include a turn-on state in which the signal light is turned on, a turn-off state in which the signal light is turned off, an ON state during which the signal light is on, and an OFF state during which the signal light is off.

2. The apparatus according to claim 1 , wherein the second classifier calculates confidence scores of the turn-on state, the turn-off state, and a steady state that implies the ON state or the OFF state.

3. The apparatus according to claim 2 , wherein the information indicates that the turn-on state is allowed to change to the turn-on state itself, the ON state, or the turn-off state, and

when the preceding state of the signal light is the turn-on state and the steady state has a higher confidence score than the turn-on state and the turn-off state, the processor identifies the state of the signal light as the ON state.

4. The apparatus according to claim 2 , wherein the information indicates that the turn-off state is allowed to change to the turn-off state itself, the OFF state, or the turn-on state, and

when the preceding state of the signal light is the turn-off state and the steady state has a higher confidence score than the turn-on state and the turn-off state, the processor identifies the state of the signal light as the OFF state.

5. The apparatus according to claim 1 , wherein the information indicates, for each of the candidate states, probabilities of transition from the candidate state to other candidate states, and

the processor calculates, for each of the candidate states, a corrected confidence score by multiplying the confidence score of the candidate state by the probability of transition from the preceding state of the signal light to the candidate state, and identifies the state of the signal light as one of the candidate states having a maximum corrected confidence score.

6. A controller for controlling travel of a vehicle, the controller comprising:

a processor configured to: input time series images into a first classifier to detect object regions each including another vehicle equipped with a signal light in the respective images, the first classifier having been trained to detect the other vehicle;

chronologically input characteristics obtained from pixel values of the object regions detected in the respective images into a second classifier to calculate confidence scores of possible candidate states of the signal light of the other vehicle, the second classifier having a recursive structure or performing a convolution operation in a temporal direction;

identify the state of the signal light, based on the preceding state of the signal light, information indicating whether transitions between the candidate states are allowed, and the confidence scores of the respective candidate states;

predict a trajectory on which the other vehicle will travel, based on the state of the signal light;

determine a trajectory to be traveled of the vehicle, based on the predicted trajectory, so that the vehicle will be separated from the other vehicle more than a predetermined distance; and

control the vehicle so that the vehicle will travel along the trajectory to be traveled.

7. A method for identifying the state of a signal light, comprising:

inputting time series images into a first classifier to detect object regions each including a vehicle equipped with a signal light in the respective images, the first classifier having been trained to detect the vehicle;

chronologically inputting characteristics obtained from pixel values of the object regions detected in the respective images into a second classifier to calculate confidence scores of possible candidate states of the signal light of the vehicle, the second classifier having a recursive structure or performing a convolution operation in a temporal direction; and

identifying the state of the signal light, based on the preceding state of the signal light, information indicating whether transitions between the candidate states are allowed, and the confidence scores of the respective candidate states;

wherein the information indicates candidate states to which the respective candidate states are allowed to change, and of the candidate states to which the preceding state of the signal light is allowed to change, the processor identifies the state of the signal light as a candidate state having a maximum confidence score; and

wherein the candidate states include a turn-on state in which the signal light is turned on, a turn-off state in which the signal light is turned off, an ON state during which the signal light is on, and an OFF state during which the signal light is off.

8. A non-transitory recording medium having recorded thereon a computer program for identifying the state of a signal light, the computer program causing a computer to execute a process comprising:

inputting time series images into a first classifier to detect object regions each including a vehicle equipped with a signal light in the respective images, the first classifier having been trained to detect the vehicle;

chronologically inputting characteristics obtained from pixel values of the object regions detected in the respective images into a second classifier to calculate confidence scores of possible candidate states of the signal light of the vehicle, the second classifier having a recursive structure or performing a convolution operation in a temporal direction; and

identifying the state of the signal light, based on the preceding state of the signal light, information indicating whether transitions between the candidate states are allowed, and the confidence scores of the respective candidate states;

wherein the information indicates candidate states to which the respective candidate states are allowed to change, and of the candidate states to which the preceding state of the signal light is allowed to change, the processor identifies the state of the signal light as a candidate state having a maximum confidence score; and

wherein the candidate states include a turn-on state in which the signal light is turned on, a turn-off state in which the signal light is turned off, an ON state during which the signal light is on, and an OFF state during which the signal light is off.